lukel97 wrote:

> I'm not sure how TuneDLenFactor2 is used by llvm, but it sounds like X100 is 
> VLEN=256 and four issue DLEN=128. While the SiFive cores that currently 
> define TuneDLenFactor2 are single issue DLEN=VLEN/2.

TuneDLenFactor2 means that DLEN = VLEN / 2, which causes instructions at LMUL N 
to be costed as if they were LMUL * 2. So e.g. <vscale x 2 x i64> is LMUL 1 for 
zvl256b. But with `-mtune=spacemit-x100` this would be costed as if it were 
LMUL 2, which roughly models one uop for each DLEN.

https://github.com/llvm/llvm-project/pull/173988
_______________________________________________
cfe-commits mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits

Reply via email to